Graph Neural Networks Beyond Homophily

Speaker:  Hanghang Tong – Urbana, USA
Topic(s):  Artificial Intelligence, Machine Learning, Computer Vision, Natural language processing

Abstract

The emergence of deep learning models designed for graph and network data, often under an umbrella term named graph neural networks (GNNs for short), has largely streamlined many graph learning problems in various science and engineering disciplines, ranging from biology, climate science, pharmaceutical science, to epidemiology. From the spectral perspective, the vast majority of the existing GNNs are essentially low-pass filters, which are based on the homophily assumption. This means that a connected node pairs tend to share the same label and/or similar embedding. However, homophily assumption is not always true where a connected or close node pairs have different labels or features (i.e., heterophily). In this talk, I will introduce some of our recent works to deal with the graph heterophily, centered around the following three research questions, including (Q1) how to automatically detect the degree of homophily?  (Q2) how to design a full-spectrum of filters? and (Q3) how to design a model-agnostic solution for handling graph heterophily?

About this Lecture

Number of Slides:  ~50
Duration:  45 minutes
Languages Available:  English
Last Updated:  18/03/2026

Request this Lecture

To request this particular lecture, please complete this online form.

Request a Tour

To request a tour with this speaker, please complete this online form.

All requests will be sent to ACM headquarters for review.